
Thinker. Builder. Artist. Exploring AI, human connections, and the art of deep conversations. Building @somoclubdotcom
Joined September 2007
- Tweets6.3K
- Following769
- Followers414
- Likes11K
Tarun retweeted
Everyone has drawn this chart in their head and is making a generational bet on which line is true.
🤖 Made with AI
Geocities 2.0 ftw
📢 NEW: Website Design Tools on Substack
Today we’re launching new website design tools that give you more control over how your Substack looks and feels.
Start with one of the preset themes in the Mini-Editor, then choose from a wide range of combinations to match your brand and reflect your style. Feeling lucky? Click the dice to try a random combo.
Want to fine-tune? The Full Editor lets you precisely shape the appearance of your Home page, Welcome page, Post pages, and more.
With these new tools, you can benefit from Substack's discovery network that helps you grow, while also building a home online that's truly yours.
25
Tarun retweeted
civilization’s amnesia loop:
1. notice a problem
2. invent a fix
3. turn the fix into tradition
4. forget what it was fixing
5. toss the tradition aside
6. rediscover the problem
Tarun retweeted
"you can aim forever, but the arrow only flies when you let go" is probably the best sentence I've read in a while
‘In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache’
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
Tarun retweeted
Oh my goodness this has to be right.
Instead of generating a random 256-bit number they generated a random number between 0 and 256.
That’s how they got 4, 66, 155, 156, 230 and 237.
Tarun retweeted
Ten years of Sad Affleck memes and it turns out he'd just been trying to install CUDA drivers the whole time.
Tarun retweeted
Gemini is the only one unilaterally pacing the frontier but none of you are giving them the kudos they deserve
Tarun retweeted
Don’t make me tap the sign
We found another cyberattack by internal OpenAI agents, this time targetting @rubygems.
They:
1) gained arbitrary remote code execution on rubydoc.
2) developed a novel exploit to steal user API keys (but we do not know if they succeeded).
They used package names including hack.rb, evil.rb, inject.rb, and exploit.rb.
We thank @j0wimo for initially discovering that agents had posted to RubyGems.
it's endearing seeing all kinds of curious people try to speedrun in one week the 2007-14 LessWrong canon with questions like "won't we just unplug it?", "can't we train AI on the US legal code?", "wouldn't it want to be good if it's smart?", and "is this some kinda Jewish plot?"
Stochastic Parrot!
Today, we’re sharing that a general-purpose internal @openai model achieved a breakthrough on one of the best-known combinatorial geometry problems. Less than 1 year ago frontier AI models were at IMO gold-level performance. I expect this pace of progress to continue.
88
this should feel surprising. in some sense it's objectively very big news. but this was so clearly the trajectory we were on that it actually just feels like another normal day. life on the METR trendline.
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.
For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.
An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.
This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
We’ve collectively decided that loneliness is a top of funnel problem.
So we built events, dinners, and meetups. All designed to answer one question: how do we get you into the room?
But you can meet a lot of people and still not have real friendships.
The IRL connection economy is a $400B+ market.
And companies are racing to own it.
In the last 6 months, $800M+ in capital was deployed on "IRL" bets.
@Tinder invested $60M into a new Events feature for connecting matches in-person. They're pivoting to IRL and offering experiences such as speakeasies, raves, and pottery classes.
@222place: raised a $10.1M Series A to curate blind social experiences for Gen Z. Personality-matched groups sent to hyperlocal nightlife events.
@JagermeisterUSA launched BestNightsVC - the only venture fund in the world dedicated solely to nightlife and IRL connection. 16 portfolio companies across 4 continents.
@timeleft: dinner with 5 strangers, every Wednesday. €18M ARR. 6,500 dinners/week across 200+ cities.
Dion: members-only social app where the first move is buying someone a real drink, redeemed IRL. 10K members, 30K+ on the waitlist founded by @revekkapal.
Pie: Bonobos founder @dunn built an IRL friendship app. $24M raised. 130K+ MAU.
@weroad_official: group trips for 20-30 year olds who don't know each other beforehand. $150M valuation.
Matchbox: is an algorithm-powered matching platform for IRL events and has powered over 100,000 connections. founded by @liamjmcgregor (prev @MarriagePact)
New dating apps like Known @Celesteamadon, Cerca @MylesCerca, and Ditto @AllenWangzian are aiming to improve connection amongst young people.
Billion-dollar companies are paying $$$ for community and events leads:
- @AnthropicAI: Marketing Events Manager ($255k)
- @tryramp: Community Manager ($223k)
- @tryramp: Events & Culture Manager ($181k)
- @duolingo: Senior Community Manager ($193k)
- @NotionHQ: Community Programs Lead
Everyone knows the more time we spend online, the more valuable real-life connection becomes.
The question isn't whether IRL wins.
It's who facilitates it best.
2
215
They optimize for the room, but they don't help you keep the people. I wrote a short post on why the traditional friendship funnel is a leaky pipe and how we’re building a "loop" instead. tarunrs.substack.com/p/the-l…
66
Tarun retweeted
Honestly, this is the most accurate diagram I've seen.
Waterfall: You plan for 18 months and deliver exactly what nobody needs anymore.
Agile: You deliver something usable at every step, but the CEO keeps asking, "Where's the car?"
AI: You get the car on day one. It has six wheels, the doors are on backwards, and it has a rocket launcher. You spend more time making it yours than actually "building"; it's shaping. owning. verifying. That's what the best AI developers do now. They don't build. They shape and own.






























